MultiBeam application for ATi GPUs released

Message boards : Number crunching : MultiBeam application for ATi GPUs released
Message board moderation

To post messages, you must log in.

Previous · 1 . . . 6 · 7 · 8 · 9 · 10 · 11 · Next

AuthorMessage
Profile Raistmer
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 16 Jun 01
Posts: 6325
Credit: 106,370,077
RAC: 121
Russia
Message 1087570 - Posted: 16 Mar 2011, 22:39:55 UTC - in response to Message 1087564.  

Both have different architectures and more suited for different kinds of tasks.
For example, HD4870 will (and AFAIK hugely) outperform NV GPUs for MilkyWay project. SETI has different kind of workload and (at least for me ;) ) it's harder to get peak performance from such kind of GPUs here.
Of course both released builds are suboptimal and could (and probably will) be optimized further (no limits for perfection, as usual ;) )
ID: 1087570 · Report as offensive
Profile Raistmer
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 16 Jun 01
Posts: 6325
Credit: 106,370,077
RAC: 121
Russia
Message 1087571 - Posted: 16 Mar 2011, 22:43:11 UTC - in response to Message 1087568.  
Last modified: 16 Mar 2011, 22:43:38 UTC

Normal & vlar's run around 50 min on my 4870, with vhar's running around 15 min. On the plus side an ATI card and run a vlar w/o choking on it.


Yes, ATI handles VLAR kind of workload better indeed.
When I tried OpenCL build on NV GPUs performance drop on VLARs was just huge.
On other ARs my GSO9600 little outperforms my HD4870 (again, both running OpenCL app, CUDA NV app still faster for NV GPUs for few known (and maybe some unknown ;D ) reasons).
ID: 1087571 · Report as offensive
Profile Sunny129
Avatar

Send message
Joined: 7 Nov 00
Posts: 190
Credit: 3,163,755
RAC: 0
United States
Message 1087573 - Posted: 16 Mar 2011, 22:45:04 UTC - in response to Message 1087564.  

Being as they are slightly faster than the GTS 250 in games
I presumed they would be atleast close with seti

Raistmer speaks the truth. if there's one thing i've learned about GPU crunching so far, its that distributed computing GPU apps are heavily dependent upon GPU architecture. hence the reason ATI cards far outperform nVidia cards in Milkyway@Home, while the exact opposite happens for other GPU apps.
ID: 1087573 · Report as offensive
Profile HAL9000
Volunteer tester
Avatar

Send message
Joined: 11 Sep 99
Posts: 6534
Credit: 196,805,888
RAC: 57
United States
Message 1087579 - Posted: 16 Mar 2011, 23:03:53 UTC - in response to Message 1087571.  

Normal & vlar's run around 50 min on my 4870, with vhar's running around 15 min. On the plus side an ATI card and run a vlar w/o choking on it.


Yes, ATI handles VLAR kind of workload better indeed.
When I tried OpenCL build on NV GPUs performance drop on VLARs was just huge.
On other ARs my GSO9600 little outperforms my HD4870 (again, both running OpenCL app, CUDA NV app still faster for NV GPUs for few known (and maybe some unknown ;D ) reasons).


Should be interesting to see how the upcoming OpenCL Intel GPU's do with the app.
SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[
ID: 1087579 · Report as offensive
aad

Send message
Joined: 3 Apr 99
Posts: 101
Credit: 204,131,099
RAC: 26
Netherlands
Message 1087594 - Posted: 16 Mar 2011, 23:57:00 UTC

My GTS250 performs equal to my HD6970 except for the VLARs.....
For Milkyway the HD6970 takes 69 seconds/wu when I let one of the CPU cores free for feeding the GPU. (and GPU running far hotter doing a Milkyway wu!)
DNETC between 14 - 15 minutes also running hot.

Temp-wise it looks like they don't run on full capacity while crunching open CL Seti. But ofcourse you already knew that!
ID: 1087594 · Report as offensive
Profile skildude
Avatar

Send message
Joined: 4 Oct 00
Posts: 9541
Credit: 50,759,529
RAC: 60
Yemen
Message 1087774 - Posted: 17 Mar 2011, 13:02:25 UTC - in response to Message 1087594.  

thats why we run 2 at a time. I haven't tried it but with 2 Gb of ram we should be able to crunch at least 3 and maybe 4 WU's at a time


In a rich man's house there is no place to spit but his face.
Diogenes Of Sinope
ID: 1087774 · Report as offensive
Profile Sunny129
Avatar

Send message
Joined: 7 Nov 00
Posts: 190
Credit: 3,163,755
RAC: 0
United States
Message 1087778 - Posted: 17 Mar 2011, 13:19:12 UTC - in response to Message 1087594.  
Last modified: 17 Mar 2011, 13:25:48 UTC

My GTS250 performs equal to my HD6970 except for the VLARs.....
For Milkyway the HD6970 takes 69 seconds/wu when I let one of the CPU cores free for feeding the GPU. (and GPU running far hotter doing a Milkyway wu!)
DNETC between 14 - 15 minutes also running hot.

Temp-wise it looks like they don't run on full capacity while crunching open CL Seti. But ofcourse you already knew that!

out of curiosity, how much slower does your 6970 crunch a MW@H task when you have zero free CPU cores (i.e. when all of your CPU cores are preoccupied by other tasks)?


thats why we run 2 at a time. I haven't tried it but with 2 Gb of ram we should be able to crunch at least 3 and maybe 4 WU's at a time

does it then follow that my 2GB version of the 5870 should also be able to crunch 3, maybe 4 MW@H tasks at once as well? or is my slightly older GPU architecture going to be a limiting factor? i ask b/c as of now, my 5870 only crunches 1 MW@H task at a time, sometimes 2 at a time, depending on my BOINC client configuration. i know i can make further edits to an .xml file to allow more tasks to run simultaneously, but i don't want to edit the code to force my GPU to take on more work than its capable of handling at once.
ID: 1087778 · Report as offensive
Cruncher-American Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor

Send message
Joined: 25 Mar 02
Posts: 1513
Credit: 370,893,186
RAC: 340
United States
Message 1087779 - Posted: 17 Mar 2011, 13:25:42 UTC

Is there a concise guide as to what one must do to use the ATI apps? I have 2 (soon to be 4) HD 5550 1GB cards (they were cheap - $25 & $30 each after rebates),
and I want to know exactly what to install for these to work properly (I have a 4 PCIe slot board for them).

Looking over this thread is quite confusing to me; if there is anything like a cookbook for this, I would appreciate being pointed to it.

Thanks!
ID: 1087779 · Report as offensive
Profile skildude
Avatar

Send message
Joined: 4 Oct 00
Posts: 9541
Credit: 50,759,529
RAC: 60
Yemen
Message 1087790 - Posted: 17 Mar 2011, 15:07:30 UTC - in response to Message 1087778.  

My GTS250 performs equal to my HD6970 except for the VLARs.....
For Milkyway the HD6970 takes 69 seconds/wu when I let one of the CPU cores free for feeding the GPU. (and GPU running far hotter doing a Milkyway wu!)
DNETC between 14 - 15 minutes also running hot.

Temp-wise it looks like they don't run on full capacity while crunching open CL Seti. But ofcourse you already knew that!

out of curiosity, how much slower does your 6970 crunch a MW@H task when you have zero free CPU cores (i.e. when all of your CPU cores are preoccupied by other tasks)?


thats why we run 2 at a time. I haven't tried it but with 2 Gb of ram we should be able to crunch at least 3 and maybe 4 WU's at a time

does it then follow that my 2GB version of the 5870 should also be able to crunch 3, maybe 4 MW@H tasks at once as well? or is my slightly older GPU architecture going to be a limiting factor? i ask b/c as of now, my 5870 only crunches 1 MW@H task at a time, sometimes 2 at a time, depending on my BOINC client configuration. i know i can make further edits to an .xml file to allow more tasks to run simultaneously, but i don't want to edit the code to force my GPU to take on more work than its capable of handling at once.

MW is not setup to run more than 1 WU at a time. you can run more than one but you'll only save loading time which is only a few seconds per WU.


In a rich man's house there is no place to spit but his face.
Diogenes Of Sinope
ID: 1087790 · Report as offensive
Profile Sunny129
Avatar

Send message
Joined: 7 Nov 00
Posts: 190
Credit: 3,163,755
RAC: 0
United States
Message 1087794 - Posted: 17 Mar 2011, 15:23:35 UTC

oops i got confused...i thought you were still talking about MW@H tasks when you mentioned running 2 or more at a time.
ID: 1087794 · Report as offensive
aad

Send message
Joined: 3 Apr 99
Posts: 101
Credit: 204,131,099
RAC: 26
Netherlands
Message 1087820 - Posted: 17 Mar 2011, 17:26:00 UTC - in response to Message 1087778.  
Last modified: 17 Mar 2011, 17:30:50 UTC

out of curiosity, how much slower does your 6970 crunch a MW@H task when you have zero free CPU cores (i.e. when all of your CPU cores are preoccupied by other tasks)?


[milkyway]
81 seconds while crunching NFS 1.09 (the 1 gig wu's!) on all 6 cores.
[/milkyway]

I never got a stable crunch while doing 2 SETIwu's at a time with this 11.2 ATI driver.
ID: 1087820 · Report as offensive
Profile Sunny129
Avatar

Send message
Joined: 7 Nov 00
Posts: 190
Credit: 3,163,755
RAC: 0
United States
Message 1087821 - Posted: 17 Mar 2011, 17:36:25 UTC - in response to Message 1087820.  

[milkyway]
81 seconds while crunching NFS 1.09 (the 1 gig wu's!) on all 6 cores.
[/milkyway]

I never got a stable crunch while doing 2 SETIwu's at a time with this 11.2 ATI driver.

interesting...thanks for the numbers. i've got a 5870 - sometimes i'm crunching S@H MB/AP with it, other times i'm crunching MW@H with it. i try to crunch as many simultaneous E@H CPU tasks as possible along side my GPU cruncher, but i have yet to see how much its slowing down my S@H and MW@H GPU tasks. looks like its time for an experiment :).
ID: 1087821 · Report as offensive
aad

Send message
Joined: 3 Apr 99
Posts: 101
Credit: 204,131,099
RAC: 26
Netherlands
Message 1091977 - Posted: 31 Mar 2011, 16:35:04 UTC

New drivers online! version 11.3
Anyone care/dare to try?
http://support.amd.com/us/kbarticles/Pages/AMDCatalyst113ReleaseNotes.aspx
ID: 1091977 · Report as offensive
aad

Send message
Joined: 3 Apr 99
Posts: 101
Credit: 204,131,099
RAC: 26
Netherlands
Message 1091998 - Posted: 31 Mar 2011, 18:27:13 UTC
Last modified: 31 Mar 2011, 18:35:14 UTC

I'll give it a try.
Got fed-up by the hangs and errors with crunching 2 wu's/time.
Deinstalled v 11.2
Rebooted
Did GPUclean
Installed v 11.3 without problems.
Rebooted
Run Boinc
Computer on website shows; CAL ATI unknown (2048MB) driver: 1.4.1332
<edit> crunching 2 AP wu's/time while monitoring with GPUZ !!! wow!
This machine http://setiathome.berkeley.edu/show_host_detail.php?hostid=5750222
ID: 1091998 · Report as offensive
Profile skildude
Avatar

Send message
Joined: 4 Oct 00
Posts: 9541
Credit: 50,759,529
RAC: 60
Yemen
Message 1092006 - Posted: 31 Mar 2011, 18:44:30 UTC - in response to Message 1091998.  

already have it onboard since last night and it appears to be fine. Only a few tweaks they mentioned the biggest is probably the WMV glitching has been resolved. Not sure any real change for the Seti folks yet though


In a rich man's house there is no place to spit but his face.
Diogenes Of Sinope
ID: 1092006 · Report as offensive
aad

Send message
Joined: 3 Apr 99
Posts: 101
Credit: 204,131,099
RAC: 26
Netherlands
Message 1092008 - Posted: 31 Mar 2011, 18:47:54 UTC

Because GPU-load is 18% with GPU-temp 28 deg. C sometimes go to max. 50% I switched to 3 wu's/time.
Still crunching happely with GPUZ enabled!
GPU-load now around 70%
GPU-Temp= 69 C with aut. fanspeed 33%
ID: 1092008 · Report as offensive
Claggy
Volunteer tester

Send message
Joined: 5 Jul 99
Posts: 4654
Credit: 47,537,079
RAC: 4
United Kingdom
Message 1092028 - Posted: 31 Mar 2011, 19:45:48 UTC - in response to Message 1091977.  
Last modified: 31 Mar 2011, 20:08:48 UTC

New drivers online! version 11.3
Anyone care/dare to try?
http://support.amd.com/us/kbarticles/Pages/AMDCatalyst113ReleaseNotes.aspx

I installed Cat 11.3 / AMD-APP-SDK-v2.4-rc1 (595.9) and 270.51 last night, my runtime on AP 0% Blanked tasks is down to 5,200 secs from 6,400 secs with Cat 11.2 :-)

Claggy
ID: 1092028 · Report as offensive
aad

Send message
Joined: 3 Apr 99
Posts: 101
Credit: 204,131,099
RAC: 26
Netherlands
Message 1092042 - Posted: 31 Mar 2011, 20:06:51 UTC


This is with 3 AP task on the GPU while doing 6 NFS Lasievef 1.09 on the CPU
Taking up all my 8 Gig RAM!
But very stable though!

Could this v 11.3 be the one?
ID: 1092042 · Report as offensive
Profile skildude
Avatar

Send message
Joined: 4 Oct 00
Posts: 9541
Credit: 50,759,529
RAC: 60
Yemen
Message 1092061 - Posted: 31 Mar 2011, 21:50:09 UTC - in response to Message 1092042.  

I tried using the gpuz and found that it didnt crash my 6970 so that's new and a bonus


In a rich man's house there is no place to spit but his face.
Diogenes Of Sinope
ID: 1092061 · Report as offensive
Profile Blurf
Volunteer tester

Send message
Joined: 2 Sep 06
Posts: 8962
Credit: 12,678,685
RAC: 0
United States
Message 1092280 - Posted: 1 Apr 2011, 19:47:36 UTC
Last modified: 1 Apr 2011, 19:58:59 UTC

Ok...installed it but it says there's no new work for anything so I can't tell if it's working or not.

Correction..just got some so we'll see how it looks after my nap


ID: 1092280 · Report as offensive
Previous · 1 . . . 6 · 7 · 8 · 9 · 10 · 11 · Next

Message boards : Number crunching : MultiBeam application for ATi GPUs released


 
©2024 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.